AFSA-RBF Neural Network Controller

Radial Radial Basis Function (RBF) Neural Network is a high-efficiencyFeedforward Neural Network. It has the characteristics of optimal approximation，global optimum，the simple construction and the fast training speed. RBF NeuralNetwork could apply on the field such as Pattern Recognition, Signal Processing,Approximation of Nonlinear Function and so on. RBF Neural Network has localadjustment and override the received domain in the simulation of the human braineach other. It also can approximate any continuous function with arbitrary precision.Artificial Fish Algorithm as a training algorithm of RBF Neural Network hascharacteristics: the good ability to overcome the local extremum and obtain theglobal extremum. There is self-adaptive ability for searching space. There is norequest for the initial value. Not sensitive for the selection of the parameters. In thispaper, we introduced the fish inventory and the optimal fish into the algorithm toimprove the efficiency of the algorithm in condition of global optimization. Applingthe micro Artificial Fish Algorithm into neural network training, the neural networktraining model based on artificial fish algorithm is built up.This paper designs the RBF Neural Network controller, containing the artificialfish algorithm and RBF Neural Network, with double inverted pendulum as theresearch object. Meanwhile, through downloading the simulated RBF model to SolidCompany’s invert pendulum system, the algorithm’s actual effect has been verifiedagain. The simulation result verifies that the designed controller satisfies theperformance requirement by simulating the double inverted pendulum in Matlab. Theexperimental results show that the double inverted pendulum can maintain stableequilibrium, good anti-interference ability, and verify the good control performanceof the RBF neural network controller in this paper finally.